[openib-general][PATCH][kdapl]: evd upcall policy implementation

[James Lentini]

We should verify that all implementations of the verbs API will behave 
this way. I'll start a new thread on the list to make sure that this 
is the correct definition.

james

On Mon, 15 Aug 2005, Guy German wrote:

guyg> Hi,
guyg> 
guyg> >> Are you suggesting that enabling the CQ upcall will not trigger the
guyg> >> CQ upcall, if completions happened before enabling?
guyg> >> I don't think this is the case, but I'm not 100% sure...
guyg> > 
guyg> > That is my assumption of how it works. This is how other
guyg> > verbs APIs have worked in the past.
guyg> 
guyg> Please see InfiniHost MT23108 Programmer's Reference Manual
guyg> p 102 section 9.4.1
guyg> 
guyg> If completions are posted to the CQ (after the reporting of a completion event) but still not consumed by the software,
guyg> events will be generated immediately after request for notification is executed. Subscribe for event is implemented by
guyg> writing the CQ doorbell with the request notification command to the appropriate UAR page, and passing as a parameter
guyg> to the command the consumer index to be polled.
guyg> 
guyg> (found by Or G.)
guyg> 
guyg> >> As I mentioned before, and regardless to this issue, I still believe
guyg> >> that the right order should be:
guyg> >>>>  call kDAPL upcall
guyg> >>>>  (conditionally) enable CQ upcall
guyg> >> We can't have interrupts if the consumer disabled the upcall
guyg> >> policy... 
guyg> > 
guyg> > I agree that we should not request interrupts if the consumer disabled
guyg> > the upcall policy.
guyg> 
guyg> Guy
guyg>